CN113642475A - Atlantic hurricane intensity estimation method based on convolutional neural network model - Google Patents

Atlantic hurricane intensity estimation method based on convolutional neural network model Download PDF

Info

Publication number
CN113642475A
CN113642475A CN202110942251.1A CN202110942251A CN113642475A CN 113642475 A CN113642475 A CN 113642475A CN 202110942251 A CN202110942251 A CN 202110942251A CN 113642475 A CN113642475 A CN 113642475A
Authority
CN
China
Prior art keywords
layer
convolution
layers
cloud
hurricane
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110942251.1A
Other languages
Chinese (zh)
Other versions
CN113642475B (en
Inventor
胡天慧
余晖
束炯
鲁小琴
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Institute Of Typhoon China Meteorological Administration
East China Normal University
Original Assignee
Shanghai Institute Of Typhoon China Meteorological Administration
East China Normal University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shanghai Institute Of Typhoon China Meteorological Administration, East China Normal University filed Critical Shanghai Institute Of Typhoon China Meteorological Administration
Priority to CN202110942251.1A priority Critical patent/CN113642475B/en
Publication of CN113642475A publication Critical patent/CN113642475A/en
Application granted granted Critical
Publication of CN113642475B publication Critical patent/CN113642475B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an Atlantic hurricane intensity estimation method based on a convolutional neural network model. The model established by the invention is that an image with the size of 114 multiplied by 114 is taken as the optimal input, 4 layers of convolutional layers and 2 layers of pooling layers are combined and then are connected with 3 layers of full connection, wherein the size of a convolution kernel of the first three layers of convolutional layers is 7 multiplied by 7, the size of a convolution kernel of the fourth layer of convolutional layers is 3 multiplied by 3, Dropout terms are introduced into the convolutional layers and the full connection layers in front and at the back, and finally, the estimated intensity is subjected to time smoothing for 18 hours, so that the finally estimated hurricane intensity is obtained. The cloud system characteristics related to tropical cyclone strength are directly extracted from the image, and compared with artificially defined characteristic factors, the cloud system characteristics are more comprehensive, the strength determining effect is better, and the cloud system characteristics have real-time performance and full-automation performance.

Description

Atlantic hurricane intensity estimation method based on convolutional neural network model
Technical Field
The invention relates to the field of tropical cyclone strength estimation (strength determination), which is to analyze and learn a large amount of satellite infrared cloud image data by utilizing a deep learning technology, automatically extract complex characteristics related to the strength of tropical cyclone cloud images from the tropical cyclone cloud images and further estimate the strength. After the constructed model is trained, the trained model can be directly used, and the accuracy of tropical cyclone strength estimation can be effectively improved.
Background
Tropical cyclones are one of the most frequently occurring, widely spread and most harmful natural disasters worldwide. Strong wind, precipitation and storm surge brought by tropical cyclone are important factors causing disasters, and the severity of the disasters is closely related to information such as positions, strength and structures of the disasters, so that the information is important for disaster defense management work by grasping the information in time.
Most of the time of the tropical cyclone is located on the ocean, the conventional meteorological observation station cannot generally cover the tropical cyclone, the remote sensing satellite can realize all-weather wide-range earth observation, and the observation data of the tropical cyclone is the main data for researching the generation development and the extinction of the tropical cyclone. Many methods for estimating the tropical cyclone strength by using satellite remote sensing observation data have been studied at home and abroad. Among the methods that have been the most well-developed and widely used globally is the Dvorak technique. The Dvorak technology is a relation between tropical cyclone intensity indexes summarized by combining a visible light infrared cloud chart with actual forecast experience and cloud system characteristic changes, and has the main defects that the subjectivity is too strong, and different analysts in business application diverge the intensity estimation of the same tropical cyclone at the same time. Over forty years, the method has been continuously improved toward automation and objectification. The microwave channel of the non-stationary satellite can penetrate through the non-strong precipitation cloud at the middle and upper layers of the tropical cyclone to detect the internal information of the tropical cyclone, so that the intensity of the tropical cyclone is reflected, but the intensity of the tropical cyclone based on the microwave data is not as strong as that of the tropical cyclone based on the infrared data due to the space-time resolution of the non-stationary satellite. With the development of artificial intelligence, various methods for deep learning have strong capabilities in solving nonlinear problems, image recognition and the like, but at present, the methods are still less applied to hurricane intensity estimation. Furthermore, the current main reason for hurricane intensity estimation accuracy is the lack of real observations, whereas there are abundant aircraft observations available in the atlantic for inspection.
In the prior art, the deep learning technology is used for automatically extracting the features of the image, and the effect of estimating the intensity of the tropical cyclone based on the features is superior to the fixed intensity effect of adopting artificially defined characteristic factors of the tropical cyclone cloud system. In the deep learning series method, the effect of strength determination by adopting various data (such as infrared channel data, microwave data and the like) is better than that by adopting single-channel data, but the acquisition of different data has time and space difference, which is not beneficial to the real-time strength determination. Meanwhile, the established deep learning model does not embody the parameter selection of modeling in detail. In addition, the main reason for restricting the strong fixing effect of tropical cyclone is lack of real observation data, most methods only compare with the best path data (non-real observation), and the invention simultaneously uses the airplane observation data (real observation) for inspection.
Disclosure of Invention
The invention aims to provide an Atlantic hurricane (tropical cyclone) intensity estimation method based on a convolutional neural network model, aiming at the technical problems to be solved.
In order to solve the technical problems, the specific technical scheme of the invention is as follows:
an Atlantic hurricane intensity estimation method based on a convolutional neural network model can automatically extract characteristic factors related to hurricane intensity by inputting cloud picture data, so as to estimate the intensity, and the method comprises the following specific steps:
step 1: acquiring an infrared bright-warm cloud picture which is matched with the national hurricane center in the last forty years in the time of the best path, removing one third of the missing image, and filling the missing image with a nearby value; finally, 35-year hurricane cloud pictures are selected for training, wherein a training set and a verification set are divided in a training sample according to a ratio of 7:3, and in addition, the hurricane cloud pictures in the rest years are used as test sets for independent inspection;
step 2: and (3) carrying out data standardization on each infrared bright and warm cloud picture, wherein the standardized formula is as follows:
Figure BDA0003215338880000021
wherein, XstandardizedIs the normalized cloud image data, x is the original cloud image data, xameanIs the average of all the cloud data, xastdThe standard deviation of all the cloud image data is obtained;
and step 3: building a deep learning environment: installing a TensorFlow library based on Python language on a server;
and 4, step 4: in the deep learning environment, optimizing a model structure of the convolutional neural network by adopting a parameter adjustment method, wherein the model structure of the convolutional neural network comprises an input layer, a convolutional layer, a pooling layer, an activation layer and a full-connection layer; the input layer, i.e. the input original data, needs to have certain length, width and depth; the convolution layer is to utilize the convolution kernel of the layer to carry out convolution operation on the input image and then obtain a characteristic graph of the layer through an activation function so as to realize the characteristic extraction of the input image; the specific calculation result of the convolution is the sum of dot products of the local image data and a convolution kernel, namely a filter, and the specific formula is as follows:
Figure BDA0003215338880000022
in the formula: l is the number of layers of the convolutional layer,
Figure BDA0003215338880000023
is the convolution kernel of the l-th layer,
Figure BDA0003215338880000024
as an offset term, MjIs an input feature;
the pooling layer is used for down-sampling the current feature map; the activation layer carries out nonlinear conversion on the characteristic information; the full connection layer converts the characteristic diagram into a one-dimensional vector;
establishing a convolutional neural network model suitable for hurricane intensity estimation, wherein the size of an input cloud picture, the number of layers of a convolutional layer and a pooling layer and the size parameter of a convolutional kernel are required to be determined; setting an initial model architecture to be 4 convolutional layers and 4 pooling layers alternately, adding 3 full-connection layers, setting the size of a convolutional core to be 3 multiplied by 3, setting the convolution step length to be 1, selecting 2 multiplied by 2 maximum pooling, selecting a Relu function as an activation function, wherein the specific formula is as follows: f (x) max (0, x); meanwhile, Mean Square Error (MSE) is selected as a loss function, and Mean Absolute Error (MAE) is selected as evaluation accuracy;
Figure BDA0003215338880000031
Figure BDA0003215338880000032
wherein n is the number of samples,
Figure BDA0003215338880000033
to predict value, yiIs the true value; training a convolutional neural network model by using a training sample, and continuously optimizing a network structure by using a parameter adjustment method;
4.1, selecting images with the sizes of 28 × 28, 84 × 84, 114 × 114, 142 × 142, 172 × 172 and 301 × 301 taking the hurricane center as the image center as input, and performing 6 tests on the rectangular images with the side lengths of 2 °, 6 °, 8 °,10 °, 12 ° and 21 ° respectively to determine the optimal input range; according to the test results, the operation was performed with the size of 114 × 114(8 ° × 8 °) as a model input;
4.2 for the feature extraction of the cloud picture, pooling operation is carried out at the lower layer of the network, cloud profile and edge feature information are effectively extracted, and the effective features are obviously lost when the pooling layer is continuously used at the higher layer of the network. And setting a combined structure of the convolutional layer and the pooling layer, and selecting an optimal network structure according to an optimal result. On the basis, convolution kernels with different sizes are arranged, the cloud image feature extraction capability of the convolution kernels is compared, and the size of the optimal convolution kernel is selected to be 7 x 7;
4.3 after the specific structure of the convolutional neural network model is determined, adding a Dropout term and optimizing the model in order to increase the robustness of the model;
through the parameter optimization of step 4.2, the network structure is set as: input layer-convolution layer 1-pooling layer 1-convolution layer 2-pooling layer 2-convolution layer 3-convolution layer 4-full-connection layer 1-full-connection layer 2-full-connection layer 3-output layer; the convolution kernel sizes of the convolution layers 1 and 2 are 7 × 7, the convolution step length is 1, the convolution kernel sizes of the convolution layers 3 and 4 are 7 × 7 and 3 × 3 respectively, and the convolution step lengths are both 2; 2 multiplied by 2 maximum pooling is selected for both the pooling layer 1 and the pooling layer 2; the number of neurons of the full junction layer 1, the full junction layer 2 and the full junction layer 3 is 512, 128 and 64 respectively; activating each convolution layer and each full-connection layer by adopting a Relu function;
and 5: and performing intensity estimation by using the trained model, and performing 18-hour time smoothing on the estimation result to obtain a final estimation intensity, wherein the final estimation intensity result has the following specific smoothing formula:
Figure BDA0003215338880000034
where Vc is the final hurricane intensity estimate, V,V6、V12、V18The estimated intensities were the current hour, the first 6 hours, the first 12 hours and the first 18 hours, respectively.
The tropical cyclone strength estimation method based on deep learning can realize the estimation of the tropical cyclone strength only by using the infrared satellite cloud picture, and has real-time property and full automation. Meanwhile, the cloud system characteristics related to the tropical cyclone strength are directly extracted from the image, and compared with artificially defined characteristic factors, the cloud system characteristics are more comprehensive and have better strength-determining effect.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a satellite infrared cloud;
FIG. 3 is a diagram illustrating training results of different input range tests;
FIG. 4 is a diagram illustrating training results for different convolution kernel sizes;
FIG. 5 is a graph of deviation distributions including Dropout for hurricane intensity estimates on test samples;
FIG. 6 is a graph of deviation distribution on test specimens after 18 hours of running average introduction;
FIG. 7 is a network architecture diagram of a convolutional neural network tropical cyclone intensity estimation;
FIG. 8 is a graph of a partial hurricane estimated intensity compared to the best path near-center maximum wind speed and ADT estimated intensity based on a convolutional neural network approach.
Detailed Description
The specific idea of the method is to utilize a convolution neural network model with each parameter set determined through experiments to extract characteristic factors related to hurricane intensity from the data of the infrared bright-temperature cloud chart of the geostationary satellite, and objectively estimate the tropical cyclone intensity. The method can realize the real-time intensity estimation of the tropical cyclone by utilizing the high space-time resolution characteristic of the satellite cloud picture data, and can provide more references for forecasters.
Examples
Referring to fig. 1, the present invention includes:
step 1: and acquiring a tropical cyclone optimal path data set, selecting an optimal path record, and screening a satellite cloud picture matched with the optimal path record. And judging the initial satellite cloud picture, directly eliminating one third of the images with missing measurement, and filling up a nearby value for a small number of missing measurements. Fig. 2(a) is a hurricane cloud with less missing measurements, and fig. 2(b) is a hurricane cloud after the correction is completed.
Step 2: the processed data is normalized as shown in fig. 2 (c).
And step 3: an image of 301 × 301 size is clipped to 114 × 114 size as a model input. The input image size is obtained by a parameter adjustment method, as shown in fig. 3.
And 4, step 4: and (4) automatically extracting the features of the cloud picture by using a convolutional neural network model, and then estimating the current intensity.
The constructed convolutional neural network, namely 4 convolutional layers are combined with 3 fully-connected layers, wherein the first 2 convolutional layers are connected with the maximum pooling, the second 2 convolutional layers are only subjected to convolution operation, and Dropout terms are added before and after the fully-connected layers, so that neurons are inactivated randomly, and the robustness of the model is increased. FIG. 4 is a graph of the effect of different convolution kernel sizes on model results, with the absolute error averaged. FIG. 5 is a graph of the percentage of error on the test samples to the total sample after the Dropout term was added, with samples having an error between [ -10kt,10kt ] accounting for 67%.
And 5: when the intensity was known in the first 18 hours, the current intensity value was smoothed to obtain the final estimated intensity value. FIG. 6 is a graph of the percentage of error on the test samples over the total sample after introducing time smoothing, where samples with an error between [ -10kt,10kt ] are improved by 5.5% over the samples before smoothing.
In conclusion, the tropical cyclone intensity estimation method based on the convolutional neural network model can realize real-time intensity estimation when the infrared cloud picture of the geostationary satellite can be obtained. Meanwhile, when the model is established, on the basis of a classical convolutional neural network architecture, the combination of a convolutional layer and a pooling layer, the size of a convolutional kernel and the influence of an overfitting phenomenon on the estimation accuracy of the tropical cyclone intensity are comprehensively considered, and the convolutional neural network optimal model suitable for estimating the tropical cyclone intensity of the Atlantic ocean is obtained. Description of the effects of the examples:
the results of the geostationary satellite infrared data and the national hurricane center optimal path data set and the available aircraft reconnaissance data were examined every 6 hour intervals by using the tropical cyclone satellite data training model in the Atlants of 1978 and 2012 and taking the tropical cyclone number in 2013 and 2016 as an example. Table 1 shows the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) of the estimates of the sums before smoothing of the test samples and the best path data. By comparison, the error of each year after the smoothing for 18 hours is reduced, and the estimation result of the integral tropical cyclone strength is obviously improved. This result also illustrates that introducing a time-sliding average in the intensity estimation can reduce to some extent the effect of sudden changes in flow activity in the short term. The average absolute error of the final independent sample and the optimal path data is 7.87kt, the root mean square error is 10.59kt, and the average absolute error and the root mean square error of the airplane reconnaissance data are 9.08kt and 11.35kt within 1 hour of the time interval.
Table 1 mean absolute error and root mean square error of test sample and best path per year and all years
Figure BDA0003215338880000051
To show the specific result of tropical cyclone strength estimation based on the convolutional neural network specifically, fig. 8 shows and compares the estimated strength of the present invention and the ADT method with four tropical cyclone life histories of wenbeot (HUMBERTO, 2013), GONZALO (GONZALO, 2014), huajin (joaqin, 2015), and GASTON (gason, 2016) as examples, and with the maximum wind speed near the center of the optimal path provided by the center of the hurricane as a reference. The analysis result shows that the convolutional neural network model can realize continuous estimation in the whole life history of the tropical cyclone, and the estimation strength is more stable.
Although it is difficult to directly compare the improvement of the tropical cyclone intensity estimation model by adopting different satellite data and different methods, the advancement of the model can be reflected to a certain extent on the overall fitting effect of the model independent test. Compared with the existing tropical cyclone strength estimation method in the Atlantic (Table 2), the error of strength estimation by adopting the deep learning method (CNN) to automatically extract the characteristics is the smallest, which shows that the method has the best effect. The application of the convolutional neural network in the estimation of the tropical cyclone strength of the Atlantic ocean better shows that the method can be extended to other sea areas.
TABLE 2 Atlantic hurricane intensity estimation based on different methods versus accuracy of best path
Figure BDA0003215338880000061

Claims (1)

1. An Atlantic hurricane intensity estimation method based on a convolutional neural network model is characterized by comprising the following specific steps:
step 1: acquiring an infrared bright-warm cloud picture which is matched with the national hurricane center in the last forty years in the time of the best path, removing one third of the missing image, and filling the missing image with a nearby value; and finally, selecting a 35-year hurricane cloud picture for training, wherein the number of the hurricane cloud pictures in the training sample is as follows, wherein the number of the hurricane cloud pictures is 7:3, dividing the training set and the verification set according to the proportion, and taking the hurricane cloud pictures of the rest years as test sets to carry out independent inspection;
step 2: and (3) carrying out data standardization on each infrared bright and warm cloud picture, wherein the standardized formula is as follows:
Figure FDA0003215338870000011
wherein, XstandardizedIs the normalized cloud image data, x is the original cloud image data, xameanIs the average of all the cloud data, xastdThe standard deviation of all the cloud image data is obtained;
and step 3: building a deep learning environment: installing a TensorFlow library based on Python language on a server;
and 4, step 4: in the deep learning environment, optimizing a model structure of the convolutional neural network by adopting a parameter adjustment method, wherein the model structure of the convolutional neural network comprises an input layer, a convolutional layer, a pooling layer, an activation layer and a full-connection layer; the input layer is input original data, and has length, width and depth; the convolution layer is to utilize the convolution kernel of the layer to carry out convolution operation on the input image and then obtain a characteristic graph of the layer through an activation function so as to realize the characteristic extraction of the input image; the specific calculation result of the convolution is the sum of dot products of the local image data and a convolution kernel, namely a filter, and the specific formula is as follows:
Figure FDA0003215338870000012
in the formula: 1 is the number of layers of the convolutional layer,
Figure FDA0003215338870000013
is the convolution kernel of layer 1 and,
Figure FDA0003215338870000014
as an offset term, MjIs an input feature;
the pooling layer is used for down-sampling the current feature map; the activation layer carries out nonlinear conversion on the characteristic information; the full connection layer converts the characteristic diagram into a one-dimensional vector;
establishing a convolutional neural network model suitable for hurricane intensity estimation, setting an initial model architecture to be 4 convolutional layers and 4 pooling layers alternately, adding 3 full-connection layers, setting the size of a convolutional core to be 3 multiplied by 3, setting the convolution step length to be 1, selecting 2 multiplied by 2 maximum pooling, and selecting a Relu function as an activation function, wherein the specific formula is as follows: f (x) max (0, x); meanwhile, Mean Square Error (MSE) is selected as a loss function, and Mean Absolute Error (MAE) is selected as evaluation accuracy;
Figure FDA0003215338870000015
Figure FDA0003215338870000016
wherein n is the number of samples,
Figure FDA0003215338870000021
to predict value, yiIs the true value; training a convolutional neural network model by using a training sample, and continuously optimizing a network structure by using a parameter adjustment method;
4.1, selecting images with the sizes of 28 × 28, 84 × 84, 114 × 114, 142 × 142, 172 × 172 and 301 × 301 taking the hurricane center as the image center as input, and performing 6 tests on the rectangular images with the side lengths of 2 °, 6 °, 8 °,10 °, 12 ° and 21 ° respectively to determine the optimal input range; according to the test results, the operation was performed with the size of 114 × 114(8 ° × 8 °) as a model input;
4.2 for the feature extraction of the cloud picture, performing pooling operation at the lower layer of the network to effectively extract cloud profile and edge feature information, setting a combined structure of a convolution layer and a pooling layer at the upper layer of the network, and selecting the size of a convolution kernel to be 7 multiplied by 7;
4.3 after the specific structure of the convolutional neural network model is determined, adding a Dropout term and optimizing the model in order to increase the robustness of the model;
through the parameter optimization of step 4.2, the network structure is set as: input layer-convolution layer 1-pooling layer 1-convolution layer 2-pooling layer 2-convolution layer 3-convolution layer 4-full-connection layer 1-full-connection layer 2-full-connection layer 3-output layer; the convolution kernel sizes of the convolution layers 1 and 2 are 7 × 7, the convolution step length is 1, the convolution kernel sizes of the convolution layers 3 and 4 are 7 × 7 and 3 × 3 respectively, and the convolution step lengths are both 2; 2 multiplied by 2 maximum pooling is selected for both the pooling layer 1 and the pooling layer 2; the number of neurons of the full junction layer 1, the full junction layer 2 and the full junction layer 3 is 512, 128 and 64 respectively; activating each convolution layer and each full-connection layer by adopting a Relu function;
and 5: and performing intensity estimation by using the trained model, and performing 18-hour time smoothing on the estimation result to obtain a final estimation intensity, wherein the final estimation intensity result has the following specific smoothing formula:
Figure FDA0003215338870000022
where Vc is the final hurricane intensity estimate, V, V6、V12、V18The estimated intensities were the current hour, the first 6 hours, the first 12 hours and the first 18 hours, respectively.
CN202110942251.1A 2021-08-17 2021-08-17 Atlantic hurricane strength estimation method based on convolutional neural network model Active CN113642475B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110942251.1A CN113642475B (en) 2021-08-17 2021-08-17 Atlantic hurricane strength estimation method based on convolutional neural network model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110942251.1A CN113642475B (en) 2021-08-17 2021-08-17 Atlantic hurricane strength estimation method based on convolutional neural network model

Publications (2)

Publication Number Publication Date
CN113642475A true CN113642475A (en) 2021-11-12
CN113642475B CN113642475B (en) 2023-04-25

Family

ID=78422372

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110942251.1A Active CN113642475B (en) 2021-08-17 2021-08-17 Atlantic hurricane strength estimation method based on convolutional neural network model

Country Status (1)

Country Link
CN (1) CN113642475B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114723997A (en) * 2022-04-29 2022-07-08 厦门大学 Composite convolution operation method based on Tropical algebra, storage medium and electronic equipment
CN115082439A (en) * 2022-07-22 2022-09-20 浙江大学 Tropical cyclone strength determining method, medium and equipment fused with satellite cloud picture space-time information
CN116977909A (en) * 2023-09-22 2023-10-31 中南民族大学 Deep learning fire intensity recognition method and system based on multi-modal data
CN114723997B (en) * 2022-04-29 2024-05-31 厦门大学 Composite convolution operation method based on graphic algebra, storage medium and electronic equipment

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104036462A (en) * 2014-06-06 2014-09-10 南京大学 Method to estimate intensity of tropical cyclone across the Northwest Pacific Ocean by using geostationary meteorological satellite image
CN110378909A (en) * 2019-06-24 2019-10-25 南京林业大学 Single wooden dividing method towards laser point cloud based on Faster R-CNN
CN111507884A (en) * 2020-04-19 2020-08-07 衡阳师范学院 Self-adaptive image steganalysis method and system based on deep convolutional neural network
AU2020102100A4 (en) * 2020-09-02 2020-10-22 Khan, Mohd. Arsh MR Disease detection using iot and machine learning in rice crops
CN112364301A (en) * 2020-11-10 2021-02-12 中国气象局上海台风研究所(上海市气象科学研究所) Slope length-based near-ground wind speed statistics downscaling correction method
CN112764129A (en) * 2021-01-22 2021-05-07 易天气(北京)科技有限公司 Method, system and terminal for thunderstorm short-term forecasting
US20210182713A1 (en) * 2019-12-16 2021-06-17 Accenture Global Solutions Limited Explainable artificial intelligence (ai) based image analytic, automatic damage detection and estimation system

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104036462A (en) * 2014-06-06 2014-09-10 南京大学 Method to estimate intensity of tropical cyclone across the Northwest Pacific Ocean by using geostationary meteorological satellite image
CN110378909A (en) * 2019-06-24 2019-10-25 南京林业大学 Single wooden dividing method towards laser point cloud based on Faster R-CNN
US20210182713A1 (en) * 2019-12-16 2021-06-17 Accenture Global Solutions Limited Explainable artificial intelligence (ai) based image analytic, automatic damage detection and estimation system
CN111507884A (en) * 2020-04-19 2020-08-07 衡阳师范学院 Self-adaptive image steganalysis method and system based on deep convolutional neural network
AU2020102100A4 (en) * 2020-09-02 2020-10-22 Khan, Mohd. Arsh MR Disease detection using iot and machine learning in rice crops
CN112364301A (en) * 2020-11-10 2021-02-12 中国气象局上海台风研究所(上海市气象科学研究所) Slope length-based near-ground wind speed statistics downscaling correction method
CN112764129A (en) * 2021-01-22 2021-05-07 易天气(北京)科技有限公司 Method, system and terminal for thunderstorm short-term forecasting

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
刘菁琪;: "浅析机器学习在气候研究中的应用" *
海滢等: "利用人工神经网络模型预测西北太平洋热带气旋生成频数" *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114723997A (en) * 2022-04-29 2022-07-08 厦门大学 Composite convolution operation method based on Tropical algebra, storage medium and electronic equipment
CN114723997B (en) * 2022-04-29 2024-05-31 厦门大学 Composite convolution operation method based on graphic algebra, storage medium and electronic equipment
CN115082439A (en) * 2022-07-22 2022-09-20 浙江大学 Tropical cyclone strength determining method, medium and equipment fused with satellite cloud picture space-time information
CN116977909A (en) * 2023-09-22 2023-10-31 中南民族大学 Deep learning fire intensity recognition method and system based on multi-modal data
CN116977909B (en) * 2023-09-22 2023-12-19 中南民族大学 Deep learning fire intensity recognition method and system based on multi-modal data

Also Published As

Publication number Publication date
CN113642475B (en) 2023-04-25

Similar Documents

Publication Publication Date Title
CN112766199B (en) Hyperspectral image classification method based on self-adaptive multi-scale feature extraction model
CN109886217B (en) Method for detecting wave height from offshore wave video based on convolutional neural network
CN113642475B (en) Atlantic hurricane strength estimation method based on convolutional neural network model
CN111666903B (en) Method for identifying thunderstorm cloud cluster in satellite cloud picture
CN104331698A (en) Remote sensing type urban image extracting method
CN113221765A (en) Vegetation phenological period extraction method based on digital camera image effective pixels
CN116449331B (en) Dust particle number concentration estimation method based on W-band radar and meteorological satellite
CN111539302B (en) Glass insulator self-explosion identification method based on multi-scale deep disturbance neural network
CN115062527A (en) Geostationary satellite sea temperature inversion method and system based on deep learning
CN108734122A (en) A kind of EO-1 hyperion city water body detection method based on adaptive samples selection
Ionescu et al. DeePS at: A deep learning model for prediction of satellite images for nowcasting purposes
CN114863361A (en) Distribution line bird nest detection method based on USRNet and YOLOv3
US7058511B2 (en) Sub-visible cloud cover assessment: VNIR-SWIR
CN109839617A (en) A kind of non-precipitation echo removing method and system of radar network composite product
CN111273376B (en) Downscaling sea surface net radiation determination method, system, equipment and storage medium
Stefanou et al. Image-derived prediction of spectral image utility for target detection applications
CN115456957B (en) Method for detecting change of remote sensing image by full-scale feature aggregation
KR101475625B1 (en) statistical signature analysis method using ocean climatic sample data by computer
CN111929680B (en) Rapid flood inundation degree evaluation method based on SAR image
CN115294398A (en) SAR image target recognition method based on multi-attitude angle joint learning
CN111624606B (en) Radar image rainfall identification method
CN114063063A (en) Geological disaster monitoring method based on synthetic aperture radar and point-like sensor
Andreev et al. Cloud detection from the Himawari-8 satellite data using a convolutional neural network
CN109726641B (en) Remote sensing image cyclic classification method based on automatic optimization of training samples
CN112613371A (en) Hyperspectral image road extraction method based on dense connection convolution neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant